Hi Pedro, Thanks for your reply,
Actually I am using 6 different sequences for scanning the same subject, so a small part of the difference could be because of randomness, but there should be a way to select the best scan from these 6 different scans. I need to know how to select the best.
any suggestion?
Regards, Sima.
2009/9/4 Pedro Paulo de Magalhães Oliveira Junior ppj@netfilter.com.br
Hi Sima, I've run 6 versions of the same scan of the same subject I got some differences too. Not so big as you found but still some differences
Probably it's the -randomness flag in the recon-all
Check: http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html cheers
Pedro Paulo de M. Oliveira Junior Diretor de Operações Netfilter & SpeedComm Telecom --- Novo Netfilter 3.2 www.Netfilter.com.br --- Novo Netfilter Small Business
2009/9/4 sima chalavi sima.chalavi@gmail.com
Dear Freesurfer experts,
We performed 6 different (pilot) structural scans from the same subject and analyzed the data using Freesurfer in order to find the best scan to be used in our real experiment.
We have checked the Freesurfer output visually and there do not seem to be any problem as described in the trouble shooting manual. So all 6 scans manage to get through the Freesurfer process just fine.
However, There are a lot of differences between the numerical results for the different scans. Please find Attached graphs of (some of ) the results of segmentation and parcellation of these 6 sequences from the statistical outputs. Now, the problem is how to select the best scan from these results.
Does any body have a standard protocol for assessing images for analysis or a standard metric, e.g. goodness of fit, from the software that we can assess without having a gold standard?
Or any other tip is also appreciated.
Thanks in advance,
Sima.
Hi Sima,
you could compute the contrast-to-noise ratio between gray and white, which will give you some idea. The overall optimization is very difficult though as there are factors like distortion, contrast uniformity, etc....
cheers Bruce
On Fri, 4 Sep 2009, sima chalavi wrote:
Hi Pedro, Thanks for your reply,
Actually I am using 6 different sequences for scanning the same subject, so a small part of the difference could be because of randomness, but there should be a way to select the best scan from these 6 different scans. I need to know how to select the best.
any suggestion?
Regards, Sima.
2009/9/4 Pedro Paulo de Magalhães Oliveira Junior ppj@netfilter.com.br
Hi Sima, I've run 6 versions of the same scan of the same subject I got some differences too. Not so big as you found but still some differences
Probably it's the -randomness flag in the recon-all
Check: http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html cheers
Pedro Paulo de M. Oliveira Junior Diretor de Operações Netfilter & SpeedComm Telecom --- Novo Netfilter 3.2 www.Netfilter.com.br --- Novo Netfilter Small Business
2009/9/4 sima chalavi sima.chalavi@gmail.com
Dear Freesurfer experts,
We performed 6 different (pilot) structural scans from the same subject and analyzed the data using Freesurfer in order to find the best scan to be used in our real experiment.
We have checked the Freesurfer output visually and there do not seem to be any problem as described in the trouble shooting manual. So all 6 scans manage to get through the Freesurfer process just fine.
However, There are a lot of differences between the numerical results for the different scans. Please find Attached graphs of (some of ) the results of segmentation and parcellation of these 6 sequences from the statistical outputs. Now, the problem is how to select the best scan from these results.
Does any body have a standard protocol for assessing images for analysis or a standard metric, e.g. goodness of fit, from the software that we can assess without having a gold standard?
Or any other tip is also appreciated.
Thanks in advance,
Sima.
Dear Bruce,
Thank you for this advice. We understand the importance of optimalising the proposed factors. Do you perhaps have tips on programs that we can use to assess these factors?
In addition, we are wondering how the factors influence the output of Freesurfer. For example, if we optimize the contrast-to-noise how will this effect the cortical thickness measure? The reason for this question is depicted in the attachments. We have compared our sequences to 'bert', lined up in talairach space. Which sequence would you prefer on the basis of visual inspection?
Thanks you again for your help, Sima.
2009/9/4 Bruce Fischl fischl@nmr.mgh.harvard.edu
Hi Sima,
you could compute the contrast-to-noise ratio between gray and white, which will give you some idea. The overall optimization is very difficult though as there are factors like distortion, contrast uniformity, etc....
cheers Bruce
On Fri, 4 Sep 2009, sima chalavi wrote:
Hi Pedro,
Thanks for your reply,
Actually I am using 6 different sequences for scanning the same subject, so a small part of the difference could be because of randomness, but there should be a way to select the best scan from these 6 different scans. I need to know how to select the best.
any suggestion?
Regards, Sima.
2009/9/4 Pedro Paulo de Magalhães Oliveira Junior ppj@netfilter.com.br
Hi Sima,
I've run 6 versions of the same scan of the same subject I got some differences too. Not so big as you found but still some differences
Probably it's the -randomness flag in the recon-all
Check: http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html < http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html
cheers
Pedro Paulo de M. Oliveira Junior Diretor de Operações Netfilter & SpeedComm Telecom --- Novo Netfilter 3.2 www.Netfilter.com.br --- Novo Netfilter Small Business
2009/9/4 sima chalavi sima.chalavi@gmail.com
Dear Freesurfer experts,
We performed 6 different (pilot) structural scans from the same subject and analyzed the data using Freesurfer in order to find the best scan to be used in our real experiment.
We have checked the Freesurfer output visually and there do not seem to be any problem as described in the trouble shooting manual. So all 6 scans manage to get through the Freesurfer process just fine.
However, There are a lot of differences between the numerical results for the different scans. Please find Attached graphs of (some of ) the results of segmentation and parcellation of these 6 sequences from the statistical outputs. Now, the problem is how to select the best scan from these results.
Does any body have a standard protocol for assessing images for analysis or a standard metric, e.g. goodness of fit, from the software that we can assess without having a gold standard?
Or any other tip is also appreciated.
Thanks in advance,
Sima.
Hi Sima,
it's hard to say just from the tiffs. A couple of things you can look at:
run mris_euler_number on the lh.orig.nofix and the rh.orig.nofix. Better sequences should result in bigger (less negative) euler numbers.
Plot the intensity distributions of the gray matter, CSF and the white matter as histograms and see how much they overlap and how broad they are.
Use mri_cnr to compute the gray/white CNR from the surfaces.
cheers, Bruce
On Mon, 14 Sep 2009, sima chalavi wrote:
Dear Bruce,
Thank you for this advice. We understand the importance of optimalising the proposed factors. Do you perhaps have tips on programs that we can use to assess these factors?
In addition, we are wondering how the factors influence the output of Freesurfer. For example, if we optimize the contrast-to-noise how will this effect the cortical thickness measure? The reason for this question is depicted in the attachments. We have compared our sequences to 'bert', lined up in talairach space. Which sequence would you prefer on the basis of visual inspection?
Thanks you again for your help, Sima.
2009/9/4 Bruce Fischl fischl@nmr.mgh.harvard.edu
Hi Sima,
you could compute the contrast-to-noise ratio between gray and white, which will give you some idea. The overall optimization is very difficult though as there are factors like distortion, contrast uniformity, etc....
cheers Bruce
On Fri, 4 Sep 2009, sima chalavi wrote:
Hi Pedro,
Thanks for your reply,
Actually I am using 6 different sequences for scanning the same subject, so a small part of the difference could be because of randomness, but there should be a way to select the best scan from these 6 different scans. I need to know how to select the best.
any suggestion?
Regards, Sima.
2009/9/4 Pedro Paulo de Magalhães Oliveira Junior ppj@netfilter.com.br
Hi Sima,
I've run 6 versions of the same scan of the same subject I got some differences too. Not so big as you found but still some differences
Probably it's the -randomness flag in the recon-all
Check: http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html < http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html
cheers
Pedro Paulo de M. Oliveira Junior Diretor de Operações Netfilter & SpeedComm Telecom --- Novo Netfilter 3.2 www.Netfilter.com.br --- Novo Netfilter Small Business
2009/9/4 sima chalavi sima.chalavi@gmail.com
Dear Freesurfer experts,
We performed 6 different (pilot) structural scans from the same subject and analyzed the data using Freesurfer in order to find the best scan to be used in our real experiment.
We have checked the Freesurfer output visually and there do not seem to be any problem as described in the trouble shooting manual. So all 6 scans manage to get through the Freesurfer process just fine.
However, There are a lot of differences between the numerical results for the different scans. Please find Attached graphs of (some of ) the results of segmentation and parcellation of these 6 sequences from the statistical outputs. Now, the problem is how to select the best scan from these results.
Does any body have a standard protocol for assessing images for analysis or a standard metric, e.g. goodness of fit, from the software that we can assess without having a gold standard?
Or any other tip is also appreciated.
Thanks in advance,
Sima.
Dear experts, I have 2 questions:
1) I created labels for some ROIs on fsavgerage (qdec analysis). I got statistics from these rois for each individual subject (by map-to-subjects, etc.), but what I am now interested in is to know the size of the ROI on the group level (so on the fsaverage surface).
Following up on the recent post, I tried 'mris_curvature_stats' but it gives error if I dont specify the subject. I am not to familiar with different commands, perhaps there is another one that would work?
2) Is there any "golden standard" for the minimum p-value or minimum size of an ROI to be considered a significant finding? I expect this may differ quite a lot, depending on how strong the effects of a variable on cortical thickness can be expected. Still, if you could comment on this or refer me to some literature, I would be very grateful.
Thank you a lot! Best regards, Aga
Try something like this
mri_segstats --slabel fsaverage lh path/to/lh.yourlabel.label --sum lh.yourlabel.sum
The 4th column of the summary file will have the Area in mm^2
doug
Agnieszka Burzynska wrote:
Dear experts, I have 2 questions:
- I created labels for some ROIs on fsavgerage (qdec analysis). I got
statistics from these rois for each individual subject (by map-to-subjects, etc.), but what I am now interested in is to know the size of the ROI on the group level (so on the fsaverage surface).
Following up on the recent post, I tried 'mris_curvature_stats' but it gives error if I dont specify the subject. I am not to familiar with different commands, perhaps there is another one that would work?
- Is there any "golden standard" for the minimum p-value or minimum size of
an ROI to be considered a significant finding? I expect this may differ quite a lot, depending on how strong the effects of a variable on cortical thickness can be expected. Still, if you could comment on this or refer me to some literature, I would be very grateful.
Thank you a lot! Best regards, Aga
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Dear Doug, Thank you for your reply! The command results in ERROR: Option --slabel unknown. I could not figure it out from the options descriptions with other option i could use, so i need some help again...
Cheers & thank you! Aga
On 9/14/09 7:48 PM, "Douglas N Greve" greve@nmr.mgh.harvard.edu wrote:
Try something like this
mri_segstats --slabel fsaverage lh path/to/lh.yourlabel.label --sum lh.yourlabel.sum
The 4th column of the summary file will have the Area in mm^2
doug
Agnieszka Burzynska wrote:
Dear experts, I have 2 questions:
- I created labels for some ROIs on fsavgerage (qdec analysis). I got
statistics from these rois for each individual subject (by map-to-subjects, etc.), but what I am now interested in is to know the size of the ROI on the group level (so on the fsaverage surface).
Following up on the recent post, I tried 'mris_curvature_stats' but it gives error if I dont specify the subject. I am not to familiar with different commands, perhaps there is another one that would work?
- Is there any "golden standard" for the minimum p-value or minimum size of
an ROI to be considered a significant finding? I expect this may differ quite a lot, depending on how strong the effects of a variable on cortical thickness can be expected. Still, if you could comment on this or refer me to some literature, I would be very grateful.
Thank you a lot! Best regards, Aga
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
You probably need a newer version then. Or you can get the size of the ROI from each subject's label (mris_anatomical_stats). That way is probably better anyway.
doug
Agnieszka Burzynska wrote:
Dear Doug, Thank you for your reply! The command results in ERROR: Option --slabel unknown. I could not figure it out from the options descriptions with other option i could use, so i need some help again...
Cheers & thank you! Aga
On 9/14/09 7:48 PM, "Douglas N Greve" greve@nmr.mgh.harvard.edu wrote:
Try something like this
mri_segstats --slabel fsaverage lh path/to/lh.yourlabel.label --sum lh.yourlabel.sum
The 4th column of the summary file will have the Area in mm^2
doug
Agnieszka Burzynska wrote:
Dear experts, I have 2 questions:
- I created labels for some ROIs on fsavgerage (qdec analysis). I got
statistics from these rois for each individual subject (by map-to-subjects, etc.), but what I am now interested in is to know the size of the ROI on the group level (so on the fsaverage surface).
Following up on the recent post, I tried 'mris_curvature_stats' but it gives error if I dont specify the subject. I am not to familiar with different commands, perhaps there is another one that would work?
- Is there any "golden standard" for the minimum p-value or minimum size of
an ROI to be considered a significant finding? I expect this may differ quite a lot, depending on how strong the effects of a variable on cortical thickness can be expected. Still, if you could comment on this or refer me to some literature, I would be very grateful.
Thank you a lot! Best regards, Aga
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Dear Bruce, Thanks so much for your email,
I ran mris_euler_number and mri_cnr (see attachment) according to your advice, it is already clear for us to remove G02 because of having too small Euler number. It is also clear that for this particular scanner the G05_H_11 sequence is the best.
For our multi-centre study we want to test more than one sequence on the next scanner. Therefore, for selecting a second best scan, I need to do further analysis. I want to test your suggestion about checking intensity distribution.
I was wondering how I retrieve the intensity values you suggested to be able to plot the intensity distribution. Secondly we were wondering how to decide the (second) best scan checking that?
Thank you again for all your help, Sima.
2009/9/14 Bruce Fischl fischl@nmr.mgh.harvard.edu
Hi Sima,
it's hard to say just from the tiffs. A couple of things you can look at:
run mris_euler_number on the lh.orig.nofix and the rh.orig.nofix. Better sequences should result in bigger (less negative) euler numbers.
Plot the intensity distributions of the gray matter, CSF and the white matter as histograms and see how much they overlap and how broad they are.
Use mri_cnr to compute the gray/white CNR from the surfaces.
cheers, Bruce
On Mon, 14 Sep 2009, sima chalavi wrote:
Dear Bruce,
Thank you for this advice. We understand the importance of optimalising the proposed factors. Do you perhaps have tips on programs that we can use to assess these factors?
In addition, we are wondering how the factors influence the output of Freesurfer. For example, if we optimize the contrast-to-noise how will this effect the cortical thickness measure? The reason for this question is depicted in the attachments. We have compared our sequences to 'bert', lined up in talairach space. Which sequence would you prefer on the basis of visual inspection?
Thanks you again for your help, Sima.
2009/9/4 Bruce Fischl fischl@nmr.mgh.harvard.edu
Hi Sima,
you could compute the contrast-to-noise ratio between gray and white, which will give you some idea. The overall optimization is very difficult though as there are factors like distortion, contrast uniformity, etc....
cheers Bruce
On Fri, 4 Sep 2009, sima chalavi wrote:
Hi Pedro,
Thanks for your reply,
Actually I am using 6 different sequences for scanning the same subject, so a small part of the difference could be because of randomness, but there should be a way to select the best scan from these 6 different scans. I need to know how to select the best.
any suggestion?
Regards, Sima.
2009/9/4 Pedro Paulo de Magalhães Oliveira Junior <ppj@netfilter.com.br
Hi Sima,
I've run 6 versions of the same scan of the same subject I got some differences too. Not so big as you found but still some differences
Probably it's the -randomness flag in the recon-all
Check:
http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html <
http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html
cheers
Pedro Paulo de M. Oliveira Junior Diretor de Operações Netfilter & SpeedComm Telecom --- Novo Netfilter 3.2 www.Netfilter.com.br --- Novo Netfilter Small Business
2009/9/4 sima chalavi sima.chalavi@gmail.com
Dear Freesurfer experts,
We performed 6 different (pilot) structural scans from the same subject and analyzed the data using Freesurfer in order to find the best scan to be used in our real experiment.
We have checked the Freesurfer output visually and there do not seem to be any problem as described in the trouble shooting manual. So all 6 scans manage to get through the Freesurfer process just fine.
However, There are a lot of differences between the numerical results for the different scans. Please find Attached graphs of (some of ) the results of segmentation and parcellation of these 6 sequences from the statistical outputs. Now, the problem is how to select the best scan from these results.
Does any body have a standard protocol for assessing images for analysis or a standard metric, e.g. goodness of fit, from the software that we can assess without having a gold standard?
Or any other tip is also appreciated.
Thanks in advance,
Sima.
Hi Sima,
you should also make sure that the surfaces are visually accurate (that is, the white and pial surfaces lie at the true boundaries). If that's the case then either 03 or one of the 05s is the next best. For the distributions, you can use the aparc+aseg.mgz to extract the correct voxel indices in matlab, then plot the distributions of the norm.mgz with hist.
cheers Bruce
On Tue, 15 Sep 2009, sima chalavi wrote:
Dear Bruce, Thanks so much for your email,
I ran mris_euler_number and mri_cnr (see attachment) according to your advice, it is already clear for us to remove G02 because of having too small Euler number. It is also clear that for this particular scanner the G05_H_11 sequence is the best.
For our multi-centre study we want to test more than one sequence on the next scanner. Therefore, for selecting a second best scan, I need to do further analysis. I want to test your suggestion about checking intensity distribution.
I was wondering how I retrieve the intensity values you suggested to be able to plot the intensity distribution. Secondly we were wondering how to decide the (second) best scan checking that?
Thank you again for all your help, Sima.
2009/9/14 Bruce Fischl fischl@nmr.mgh.harvard.edu
Hi Sima,
it's hard to say just from the tiffs. A couple of things you can look at:
run mris_euler_number on the lh.orig.nofix and the rh.orig.nofix. Better sequences should result in bigger (less negative) euler numbers.
Plot the intensity distributions of the gray matter, CSF and the white matter as histograms and see how much they overlap and how broad they are.
Use mri_cnr to compute the gray/white CNR from the surfaces.
cheers, Bruce
On Mon, 14 Sep 2009, sima chalavi wrote:
Dear Bruce,
Thank you for this advice. We understand the importance of optimalising the proposed factors. Do you perhaps have tips on programs that we can use to assess these factors?
In addition, we are wondering how the factors influence the output of Freesurfer. For example, if we optimize the contrast-to-noise how will this effect the cortical thickness measure? The reason for this question is depicted in the attachments. We have compared our sequences to 'bert', lined up in talairach space. Which sequence would you prefer on the basis of visual inspection?
Thanks you again for your help, Sima.
2009/9/4 Bruce Fischl fischl@nmr.mgh.harvard.edu
Hi Sima,
you could compute the contrast-to-noise ratio between gray and white, which will give you some idea. The overall optimization is very difficult though as there are factors like distortion, contrast uniformity, etc....
cheers Bruce
On Fri, 4 Sep 2009, sima chalavi wrote:
Hi Pedro,
Thanks for your reply,
Actually I am using 6 different sequences for scanning the same subject, so a small part of the difference could be because of randomness, but there should be a way to select the best scan from these 6 different scans. I need to know how to select the best.
any suggestion?
Regards, Sima.
2009/9/4 Pedro Paulo de Magalhães Oliveira Junior <ppj@netfilter.com.br
Hi Sima,
I've run 6 versions of the same scan of the same subject I got some differences too. Not so big as you found but still some differences
Probably it's the -randomness flag in the recon-all
Check:
http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html <
http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg11235.html
>
> cheers
Pedro Paulo de M. Oliveira Junior Diretor de Operações Netfilter & SpeedComm Telecom --- Novo Netfilter 3.2 www.Netfilter.com.br --- Novo Netfilter Small Business
2009/9/4 sima chalavi sima.chalavi@gmail.com
Dear Freesurfer experts, > > We performed 6 different (pilot) structural scans from the same > subject > and analyzed the data using Freesurfer in order to find the best scan > to > be used in our real experiment. > > We have checked the Freesurfer output visually and there do not seem > to > be > any problem as described in the trouble shooting manual. So all 6 > scans > manage to get through the Freesurfer process just fine. > > > However, There are a lot of differences between the numerical results > for > the different scans. Please find Attached graphs of (some of ) the > results > of segmentation and parcellation of these 6 sequences from the > statistical > outputs. > Now, the problem is how to select the best scan from these results. > > Does any body have a standard protocol for assessing images for > analysis > or a standard metric, e.g. goodness of fit, from the software that we > can > assess without having a gold standard? > > Or any other tip is also appreciated. > > Thanks in advance, > > Sima. > > > >
freesurfer@nmr.mgh.harvard.edu
-
Agnieszka Burzynska -
Bruce Fischl -
Douglas N Greve -
sima chalavi